drivers/xen/events/events_base.c

   1 /*
   2  * Xen event channels
   3  *
   4  * Xen models interrupts with abstract event channels.  Because each
   5  * domain gets 1024 event channels, but NR_IRQ is not that large, we
   6  * must dynamically map irqs<->event channels.  The event channels
   7  * interface with the rest of the kernel by defining a xen interrupt
   8  * chip.  When an event is received, it is mapped to an irq and sent
   9  * through the normal interrupt processing path.
  10  *
  11  * There are four kinds of events which can be mapped to an event
  12  * channel:
  13  *
  14  * 1. Inter-domain notifications.  This includes all the virtual
  15  *    device events, since they're driven by front-ends in another domain
  16  *    (typically dom0).
  17  * 2. VIRQs, typically used for timers.  These are per-cpu events.
  18  * 3. IPIs.
  19  * 4. PIRQs - Hardware interrupts.
  20  *
  21  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
  22  */
  23
  24 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
  25
  26 #include <linux/linkage.h>
  27 #include <linux/interrupt.h>
  28 #include <linux/irq.h>
  29 #include <linux/module.h>
  30 #include <linux/string.h>
  31 #include <linux/bootmem.h>
  32 #include <linux/slab.h>
  33 #include <linux/irqnr.h>
  34 #include <linux/pci.h>
  35
  36 #ifdef CONFIG_X86
  37 #include <asm/desc.h>
  38 #include <asm/ptrace.h>
  39 #include <asm/irq.h>
  40 #include <asm/idle.h>
  41 #include <asm/io_apic.h>
  42 #include <asm/xen/page.h>
  43 #include <asm/xen/pci.h>
  44 #endif
  45 #include <asm/sync_bitops.h>
  46 #include <asm/xen/hypercall.h>
  47 #include <asm/xen/hypervisor.h>
  48
  49 #include <xen/xen.h>
  50 #include <xen/hvm.h>
  51 #include <xen/xen-ops.h>
  52 #include <xen/events.h>
  53 #include <xen/interface/xen.h>
  54 #include <xen/interface/event_channel.h>
  55 #include <xen/interface/hvm/hvm_op.h>
  56 #include <xen/interface/hvm/params.h>
  57 #include <xen/interface/physdev.h>
  58 #include <xen/interface/sched.h>
  59 #include <xen/interface/vcpu.h>
  60 #include <asm/hw_irq.h>
  61
  62 #include "events_internal.h"
  63
  64 const struct evtchn_ops *evtchn_ops;
  65
  66 /*
  67  * This lock protects updates to the following mapping and reference-count
  68  * arrays. The lock does not need to be acquired to read the mapping tables.
  69  */
  70 static DEFINE_MUTEX(irq_mapping_update_lock);
  71
  72 static LIST_HEAD(xen_irq_list_head);
  73
  74 /* IRQ <-> VIRQ mapping. */
  75 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
  76
  77 /* IRQ <-> IPI mapping */
  78 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
  79
  80 int **evtchn_to_irq;
  81 #ifdef CONFIG_X86
  82 static unsigned long *pirq_eoi_map;
  83 #endif
  84 static bool (*pirq_needs_eoi)(unsigned irq);
  85
  86 #define EVTCHN_ROW(e)  (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
  87 #define EVTCHN_COL(e)  (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
  88 #define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
  89
  90 /* Xen will never allocate port zero for any purpose. */
  91 #define VALID_EVTCHN(chn)       ((chn) != 0)
  92
  93 static struct irq_chip xen_dynamic_chip;
  94 static struct irq_chip xen_percpu_chip;
  95 static struct irq_chip xen_pirq_chip;
  96 static void enable_dynirq(struct irq_data *data);
  97 static void disable_dynirq(struct irq_data *data);
  98
  99 static void clear_evtchn_to_irq_row(unsigned row)
 100 {
 101         unsigned col;
 102
 103         for (col = 0; col < EVTCHN_PER_ROW; col++)
 104                 evtchn_to_irq[row][col] = -1;
 105 }
 106
 107 static void clear_evtchn_to_irq_all(void)
 108 {
 109         unsigned row;
 110
 111         for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
 112                 if (evtchn_to_irq[row] == NULL)
 113                         continue;
 114                 clear_evtchn_to_irq_row(row);
 115         }
 116 }
 117
 118 static int set_evtchn_to_irq(unsigned evtchn, unsigned irq)
 119 {
 120         unsigned row;
 121         unsigned col;
 122
 123         if (evtchn >= xen_evtchn_max_channels())
 124                 return -EINVAL;
 125
 126         row = EVTCHN_ROW(evtchn);
 127         col = EVTCHN_COL(evtchn);
 128
 129         if (evtchn_to_irq[row] == NULL) {
 130                 /* Unallocated irq entries return -1 anyway */
 131                 if (irq == -1)
 132                         return 0;
 133
 134                 evtchn_to_irq[row] = (int *)get_zeroed_page(GFP_KERNEL);
 135                 if (evtchn_to_irq[row] == NULL)
 136                         return -ENOMEM;
 137
 138                 clear_evtchn_to_irq_row(row);
 139         }
 140
 141         evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)] = irq;
 142         return 0;
 143 }
 144
 145 int get_evtchn_to_irq(unsigned evtchn)
 146 {
 147         if (evtchn >= xen_evtchn_max_channels())
 148                 return -1;
 149         if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
 150                 return -1;
 151         return evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)];
 152 }
 153
 154 /* Get info for IRQ */
 155 struct irq_info *info_for_irq(unsigned irq)
 156 {
 157         return irq_get_handler_data(irq);
 158 }
 159
 160 /* Constructors for packed IRQ information. */
 161 static int xen_irq_info_common_setup(struct irq_info *info,
 162                                      unsigned irq,
 163                                      enum xen_irq_type type,
 164                                      unsigned evtchn,
 165                                      unsigned short cpu)
 166 {
 167         int ret;
 168
 169         BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
 170
 171         info->type = type;
 172         info->irq = irq;
 173         info->evtchn = evtchn;
 174         info->cpu = cpu;
 175
 176         ret = set_evtchn_to_irq(evtchn, irq);
 177         if (ret < 0)
 178                 return ret;
 179
 180         irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
 181
 182         return xen_evtchn_port_setup(info);
 183 }
 184
 185 static int xen_irq_info_evtchn_setup(unsigned irq,
 186                                      unsigned evtchn)
 187 {
 188         struct irq_info *info = info_for_irq(irq);
 189
 190         return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
 191 }
 192
 193 static int xen_irq_info_ipi_setup(unsigned cpu,
 194                                   unsigned irq,
 195                                   unsigned evtchn,
 196                                   enum ipi_vector ipi)
 197 {
 198         struct irq_info *info = info_for_irq(irq);
 199
 200         info->u.ipi = ipi;
 201
 202         per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 203
 204         return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
 205 }
 206
 207 static int xen_irq_info_virq_setup(unsigned cpu,
 208                                    unsigned irq,
 209                                    unsigned evtchn,
 210                                    unsigned virq)
 211 {
 212         struct irq_info *info = info_for_irq(irq);
 213
 214         info->u.virq = virq;
 215
 216         per_cpu(virq_to_irq, cpu)[virq] = irq;
 217
 218         return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
 219 }
 220
 221 static int xen_irq_info_pirq_setup(unsigned irq,
 222                                    unsigned evtchn,
 223                                    unsigned pirq,
 224                                    unsigned gsi,
 225                                    uint16_t domid,
 226                                    unsigned char flags)
 227 {
 228         struct irq_info *info = info_for_irq(irq);
 229
 230         info->u.pirq.pirq = pirq;
 231         info->u.pirq.gsi = gsi;
 232         info->u.pirq.domid = domid;
 233         info->u.pirq.flags = flags;
 234
 235         return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
 236 }
 237
 238 static void xen_irq_info_cleanup(struct irq_info *info)
 239 {
 240         set_evtchn_to_irq(info->evtchn, -1);
 241         info->evtchn = 0;
 242 }
 243
 244 /*
 245  * Accessors for packed IRQ information.
 246  */
 247 unsigned int evtchn_from_irq(unsigned irq)
 248 {
 249         if (unlikely(WARN(irq >= nr_irqs, "Invalid irq %d!\n", irq)))
 250                 return 0;
 251
 252         return info_for_irq(irq)->evtchn;
 253 }
 254
 255 unsigned irq_from_evtchn(unsigned int evtchn)
 256 {
 257         return get_evtchn_to_irq(evtchn);
 258 }
 259 EXPORT_SYMBOL_GPL(irq_from_evtchn);
 260
 261 int irq_from_virq(unsigned int cpu, unsigned int virq)
 262 {
 263         return per_cpu(virq_to_irq, cpu)[virq];
 264 }
 265
 266 static enum ipi_vector ipi_from_irq(unsigned irq)
 267 {
 268         struct irq_info *info = info_for_irq(irq);
 269
 270         BUG_ON(info == NULL);
 271         BUG_ON(info->type != IRQT_IPI);
 272
 273         return info->u.ipi;
 274 }
 275
 276 static unsigned virq_from_irq(unsigned irq)
 277 {
 278         struct irq_info *info = info_for_irq(irq);
 279
 280         BUG_ON(info == NULL);
 281         BUG_ON(info->type != IRQT_VIRQ);
 282
 283         return info->u.virq;
 284 }
 285
 286 static unsigned pirq_from_irq(unsigned irq)
 287 {
 288         struct irq_info *info = info_for_irq(irq);
 289
 290         BUG_ON(info == NULL);
 291         BUG_ON(info->type != IRQT_PIRQ);
 292
 293         return info->u.pirq.pirq;
 294 }
 295
 296 static enum xen_irq_type type_from_irq(unsigned irq)
 297 {
 298         return info_for_irq(irq)->type;
 299 }
 300
 301 unsigned cpu_from_irq(unsigned irq)
 302 {
 303         return info_for_irq(irq)->cpu;
 304 }
 305
 306 unsigned int cpu_from_evtchn(unsigned int evtchn)
 307 {
 308         int irq = get_evtchn_to_irq(evtchn);
 309         unsigned ret = 0;
 310
 311         if (irq != -1)
 312                 ret = cpu_from_irq(irq);
 313
 314         return ret;
 315 }
 316
 317 #ifdef CONFIG_X86
 318 static bool pirq_check_eoi_map(unsigned irq)
 319 {
 320         return test_bit(pirq_from_irq(irq), pirq_eoi_map);
 321 }
 322 #endif
 323
 324 static bool pirq_needs_eoi_flag(unsigned irq)
 325 {
 326         struct irq_info *info = info_for_irq(irq);
 327         BUG_ON(info->type != IRQT_PIRQ);
 328
 329         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 330 }
 331
 332 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 333 {
 334         int irq = get_evtchn_to_irq(chn);
 335         struct irq_info *info = info_for_irq(irq);
 336
 337         BUG_ON(irq == -1);
 338 #ifdef CONFIG_SMP
 339         cpumask_copy(irq_get_irq_data(irq)->affinity, cpumask_of(cpu));
 340 #endif
 341         xen_evtchn_port_bind_to_cpu(info, cpu);
 342
 343         info->cpu = cpu;
 344 }
 345
 346 static void xen_evtchn_mask_all(void)
 347 {
 348         unsigned int evtchn;
 349
 350         for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
 351                 mask_evtchn(evtchn);
 352 }
 353
 354 /**
 355  * notify_remote_via_irq - send event to remote end of event channel via irq
 356  * @irq: irq of event channel to send event to
 357  *
 358  * Unlike notify_remote_via_evtchn(), this is safe to use across
 359  * save/restore. Notifications on a broken connection are silently
 360  * dropped.
 361  */
 362 void notify_remote_via_irq(int irq)
 363 {
 364         int evtchn = evtchn_from_irq(irq);
 365
 366         if (VALID_EVTCHN(evtchn))
 367                 notify_remote_via_evtchn(evtchn);
 368 }
 369 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 370
 371 static void xen_irq_init(unsigned irq)
 372 {
 373         struct irq_info *info;
 374 #ifdef CONFIG_SMP
 375         /* By default all event channels notify CPU#0. */
 376         cpumask_copy(irq_get_irq_data(irq)->affinity, cpumask_of(0));
 377 #endif
 378
 379         info = kzalloc(sizeof(*info), GFP_KERNEL);
 380         if (info == NULL)
 381                 panic("Unable to allocate metadata for IRQ%d\n", irq);
 382
 383         info->type = IRQT_UNBOUND;
 384         info->refcnt = -1;
 385
 386         irq_set_handler_data(irq, info);
 387
 388         list_add_tail(&info->list, &xen_irq_list_head);
 389 }
 390
 391 static int __must_check xen_allocate_irqs_dynamic(int nvec)
 392 {
 393         int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
 394
 395         if (irq >= 0) {
 396                 for (i = 0; i < nvec; i++)
 397                         xen_irq_init(irq + i);
 398         }
 399
 400         return irq;
 401 }
 402
 403 static inline int __must_check xen_allocate_irq_dynamic(void)
 404 {
 405
 406         return xen_allocate_irqs_dynamic(1);
 407 }
 408
 409 static int __must_check xen_allocate_irq_gsi(unsigned gsi)
 410 {
 411         int irq;
 412
 413         /*
 414          * A PV guest has no concept of a GSI (since it has no ACPI
 415          * nor access to/knowledge of the physical APICs). Therefore
 416          * all IRQs are dynamically allocated from the entire IRQ
 417          * space.
 418          */
 419         if (xen_pv_domain() && !xen_initial_domain())
 420                 return xen_allocate_irq_dynamic();
 421
 422         /* Legacy IRQ descriptors are already allocated by the arch. */
 423         if (gsi < NR_IRQS_LEGACY)
 424                 irq = gsi;
 425         else
 426                 irq = irq_alloc_desc_at(gsi, -1);
 427
 428         xen_irq_init(irq);
 429
 430         return irq;
 431 }
 432
 433 static void xen_free_irq(unsigned irq)
 434 {
 435         struct irq_info *info = irq_get_handler_data(irq);
 436
 437         if (WARN_ON(!info))
 438                 return;
 439
 440         list_del(&info->list);
 441
 442         irq_set_handler_data(irq, NULL);
 443
 444         WARN_ON(info->refcnt > 0);
 445
 446         kfree(info);
 447
 448         /* Legacy IRQ descriptors are managed by the arch. */
 449         if (irq < NR_IRQS_LEGACY)
 450                 return;
 451
 452         irq_free_desc(irq);
 453 }
 454
 455 static void xen_evtchn_close(unsigned int port)
 456 {
 457         struct evtchn_close close;
 458
 459         close.port = port;
 460         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 461                 BUG();
 462 }
 463
 464 static void pirq_query_unmask(int irq)
 465 {
 466         struct physdev_irq_status_query irq_status;
 467         struct irq_info *info = info_for_irq(irq);
 468
 469         BUG_ON(info->type != IRQT_PIRQ);
 470
 471         irq_status.irq = pirq_from_irq(irq);
 472         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 473                 irq_status.flags = 0;
 474
 475         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 476         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 477                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 478 }
 479
 480 static void eoi_pirq(struct irq_data *data)
 481 {
 482         int evtchn = evtchn_from_irq(data->irq);
 483         struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
 484         int rc = 0;
 485
 486         irq_move_irq(data);
 487
 488         if (VALID_EVTCHN(evtchn))
 489                 clear_evtchn(evtchn);
 490
 491         if (pirq_needs_eoi(data->irq)) {
 492                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 493                 WARN_ON(rc);
 494         }
 495 }
 496
 497 static void mask_ack_pirq(struct irq_data *data)
 498 {
 499         disable_dynirq(data);
 500         eoi_pirq(data);
 501 }
 502
 503 static unsigned int __startup_pirq(unsigned int irq)
 504 {
 505         struct evtchn_bind_pirq bind_pirq;
 506         struct irq_info *info = info_for_irq(irq);
 507         int evtchn = evtchn_from_irq(irq);
 508         int rc;
 509
 510         BUG_ON(info->type != IRQT_PIRQ);
 511
 512         if (VALID_EVTCHN(evtchn))
 513                 goto out;
 514
 515         bind_pirq.pirq = pirq_from_irq(irq);
 516         /* NB. We are happy to share unless we are probing. */
 517         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 518                                         BIND_PIRQ__WILL_SHARE : 0;
 519         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 520         if (rc != 0) {
 521                 pr_warn("Failed to obtain physical IRQ %d\n", irq);
 522                 return 0;
 523         }
 524         evtchn = bind_pirq.port;
 525
 526         pirq_query_unmask(irq);
 527
 528         rc = set_evtchn_to_irq(evtchn, irq);
 529         if (rc != 0) {
 530                 pr_err("irq%d: Failed to set port to irq mapping (%d)\n",
 531                        irq, rc);
 532                 xen_evtchn_close(evtchn);
 533                 return 0;
 534         }
 535         bind_evtchn_to_cpu(evtchn, 0);
 536         info->evtchn = evtchn;
 537
 538 out:
 539         unmask_evtchn(evtchn);
 540         eoi_pirq(irq_get_irq_data(irq));
 541
 542         return 0;
 543 }
 544
 545 static unsigned int startup_pirq(struct irq_data *data)
 546 {
 547         return __startup_pirq(data->irq);
 548 }
 549
 550 static void shutdown_pirq(struct irq_data *data)
 551 {
 552         unsigned int irq = data->irq;
 553         struct irq_info *info = info_for_irq(irq);
 554         unsigned evtchn = evtchn_from_irq(irq);
 555
 556         BUG_ON(info->type != IRQT_PIRQ);
 557
 558         if (!VALID_EVTCHN(evtchn))
 559                 return;
 560
 561         mask_evtchn(evtchn);
 562         xen_evtchn_close(evtchn);
 563         xen_irq_info_cleanup(info);
 564 }
 565
 566 static void enable_pirq(struct irq_data *data)
 567 {
 568         startup_pirq(data);
 569 }
 570
 571 static void disable_pirq(struct irq_data *data)
 572 {
 573         disable_dynirq(data);
 574 }
 575
 576 int xen_irq_from_gsi(unsigned gsi)
 577 {
 578         struct irq_info *info;
 579
 580         list_for_each_entry(info, &xen_irq_list_head, list) {
 581                 if (info->type != IRQT_PIRQ)
 582                         continue;
 583
 584                 if (info->u.pirq.gsi == gsi)
 585                         return info->irq;
 586         }
 587
 588         return -1;
 589 }
 590 EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
 591
 592 static void __unbind_from_irq(unsigned int irq)
 593 {
 594         int evtchn = evtchn_from_irq(irq);
 595         struct irq_info *info = irq_get_handler_data(irq);
 596
 597         if (info->refcnt > 0) {
 598                 info->refcnt--;
 599                 if (info->refcnt != 0)
 600                         return;
 601         }
 602
 603         if (VALID_EVTCHN(evtchn)) {
 604                 unsigned int cpu = cpu_from_irq(irq);
 605
 606                 xen_evtchn_close(evtchn);
 607
 608                 switch (type_from_irq(irq)) {
 609                 case IRQT_VIRQ:
 610                         per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
 611                         break;
 612                 case IRQT_IPI:
 613                         per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
 614                         break;
 615                 default:
 616                         break;
 617                 }
 618
 619                 xen_irq_info_cleanup(info);
 620         }
 621
 622         BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
 623
 624         xen_free_irq(irq);
 625 }
 626
 627 /*
 628  * Do not make any assumptions regarding the relationship between the
 629  * IRQ number returned here and the Xen pirq argument.
 630  *
 631  * Note: We don't assign an event channel until the irq actually started
 632  * up.  Return an existing irq if we've already got one for the gsi.
 633  *
 634  * Shareable implies level triggered, not shareable implies edge
 635  * triggered here.
 636  */
 637 int xen_bind_pirq_gsi_to_irq(unsigned gsi,
 638                              unsigned pirq, int shareable, char *name)
 639 {
 640         int irq = -1;
 641         struct physdev_irq irq_op;
 642         int ret;
 643
 644         mutex_lock(&irq_mapping_update_lock);
 645
 646         irq = xen_irq_from_gsi(gsi);
 647         if (irq != -1) {
 648                 pr_info("%s: returning irq %d for gsi %u\n",
 649                         __func__, irq, gsi);
 650                 goto out;
 651         }
 652
 653         irq = xen_allocate_irq_gsi(gsi);
 654         if (irq < 0)
 655                 goto out;
 656
 657         irq_op.irq = irq;
 658         irq_op.vector = 0;
 659
 660         /* Only the privileged domain can do this. For non-priv, the pcifront
 661          * driver provides a PCI bus that does the call to do exactly
 662          * this in the priv domain. */
 663         if (xen_initial_domain() &&
 664             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 665                 xen_free_irq(irq);
 666                 irq = -ENOSPC;
 667                 goto out;
 668         }
 669
 670         ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
 671                                shareable ? PIRQ_SHAREABLE : 0);
 672         if (ret < 0) {
 673                 __unbind_from_irq(irq);
 674                 irq = ret;
 675                 goto out;
 676         }
 677
 678         pirq_query_unmask(irq);
 679         /* We try to use the handler with the appropriate semantic for the
 680          * type of interrupt: if the interrupt is an edge triggered
 681          * interrupt we use handle_edge_irq.
 682          *
 683          * On the other hand if the interrupt is level triggered we use
 684          * handle_fasteoi_irq like the native code does for this kind of
 685          * interrupts.
 686          *
 687          * Depending on the Xen version, pirq_needs_eoi might return true
 688          * not only for level triggered interrupts but for edge triggered
 689          * interrupts too. In any case Xen always honors the eoi mechanism,
 690          * not injecting any more pirqs of the same kind if the first one
 691          * hasn't received an eoi yet. Therefore using the fasteoi handler
 692          * is the right choice either way.
 693          */
 694         if (shareable)
 695                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 696                                 handle_fasteoi_irq, name);
 697         else
 698                 irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 699                                 handle_edge_irq, name);
 700
 701 out:
 702         mutex_unlock(&irq_mapping_update_lock);
 703
 704         return irq;
 705 }
 706
 707 #ifdef CONFIG_PCI_MSI
 708 int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
 709 {
 710         int rc;
 711         struct physdev_get_free_pirq op_get_free_pirq;
 712
 713         op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
 714         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 715
 716         WARN_ONCE(rc == -ENOSYS,
 717                   "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
 718
 719         return rc ? -1 : op_get_free_pirq.pirq;
 720 }
 721
 722 int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
 723                              int pirq, int nvec, const char *name, domid_t domid)
 724 {
 725         int i, irq, ret;
 726
 727         mutex_lock(&irq_mapping_update_lock);
 728
 729         irq = xen_allocate_irqs_dynamic(nvec);
 730         if (irq < 0)
 731                 goto out;
 732
 733         for (i = 0; i < nvec; i++) {
 734                 irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
 735
 736                 ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
 737                                               i == 0 ? 0 : PIRQ_MSI_GROUP);
 738                 if (ret < 0)
 739                         goto error_irq;
 740         }
 741
 742         ret = irq_set_msi_desc(irq, msidesc);
 743         if (ret < 0)
 744                 goto error_irq;
 745 out:
 746         mutex_unlock(&irq_mapping_update_lock);
 747         return irq;
 748 error_irq:
 749         for (; i >= 0; i--)
 750                 __unbind_from_irq(irq + i);
 751         mutex_unlock(&irq_mapping_update_lock);
 752         return ret;
 753 }
 754 #endif
 755
 756 int xen_destroy_irq(int irq)
 757 {
 758         struct physdev_unmap_pirq unmap_irq;
 759         struct irq_info *info = info_for_irq(irq);
 760         int rc = -ENOENT;
 761
 762         mutex_lock(&irq_mapping_update_lock);
 763
 764         /*
 765          * If trying to remove a vector in a MSI group different
 766          * than the first one skip the PIRQ unmap unless this vector
 767          * is the first one in the group.
 768          */
 769         if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
 770                 unmap_irq.pirq = info->u.pirq.pirq;
 771                 unmap_irq.domid = info->u.pirq.domid;
 772                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 773                 /* If another domain quits without making the pci_disable_msix
 774                  * call, the Xen hypervisor takes care of freeing the PIRQs
 775                  * (free_domain_pirqs).
 776                  */
 777                 if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
 778                         pr_info("domain %d does not have %d anymore\n",
 779                                 info->u.pirq.domid, info->u.pirq.pirq);
 780                 else if (rc) {
 781                         pr_warn("unmap irq failed %d\n", rc);
 782                         goto out;
 783                 }
 784         }
 785
 786         xen_free_irq(irq);
 787
 788 out:
 789         mutex_unlock(&irq_mapping_update_lock);
 790         return rc;
 791 }
 792
 793 int xen_irq_from_pirq(unsigned pirq)
 794 {
 795         int irq;
 796
 797         struct irq_info *info;
 798
 799         mutex_lock(&irq_mapping_update_lock);
 800
 801         list_for_each_entry(info, &xen_irq_list_head, list) {
 802                 if (info->type != IRQT_PIRQ)
 803                         continue;
 804                 irq = info->irq;
 805                 if (info->u.pirq.pirq == pirq)
 806                         goto out;
 807         }
 808         irq = -1;
 809 out:
 810         mutex_unlock(&irq_mapping_update_lock);
 811
 812         return irq;
 813 }
 814
 815
 816 int xen_pirq_from_irq(unsigned irq)
 817 {
 818         return pirq_from_irq(irq);
 819 }
 820 EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
 821
 822 int bind_evtchn_to_irq(unsigned int evtchn)
 823 {
 824         int irq;
 825         int ret;
 826
 827         if (evtchn >= xen_evtchn_max_channels())
 828                 return -ENOMEM;
 829
 830         mutex_lock(&irq_mapping_update_lock);
 831
 832         irq = get_evtchn_to_irq(evtchn);
 833
 834         if (irq == -1) {
 835                 irq = xen_allocate_irq_dynamic();
 836                 if (irq < 0)
 837                         goto out;
 838
 839                 irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 840                                               handle_edge_irq, "event");
 841
 842                 ret = xen_irq_info_evtchn_setup(irq, evtchn);
 843                 if (ret < 0) {
 844                         __unbind_from_irq(irq);
 845                         irq = ret;
 846                         goto out;
 847                 }
 848                 /* New interdomain events are bound to VCPU 0. */
 849                 bind_evtchn_to_cpu(evtchn, 0);
 850         } else {
 851                 struct irq_info *info = info_for_irq(irq);
 852                 WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
 853         }
 854
 855 out:
 856         mutex_unlock(&irq_mapping_update_lock);
 857
 858         return irq;
 859 }
 860 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 861
 862 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 863 {
 864         struct evtchn_bind_ipi bind_ipi;
 865         int evtchn, irq;
 866         int ret;
 867
 868         mutex_lock(&irq_mapping_update_lock);
 869
 870         irq = per_cpu(ipi_to_irq, cpu)[ipi];
 871
 872         if (irq == -1) {
 873                 irq = xen_allocate_irq_dynamic();
 874                 if (irq < 0)
 875                         goto out;
 876
 877                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 878                                               handle_percpu_irq, "ipi");
 879
 880                 bind_ipi.vcpu = cpu;
 881                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 882                                                 &bind_ipi) != 0)
 883                         BUG();
 884                 evtchn = bind_ipi.port;
 885
 886                 ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
 887                 if (ret < 0) {
 888                         __unbind_from_irq(irq);
 889                         irq = ret;
 890                         goto out;
 891                 }
 892                 bind_evtchn_to_cpu(evtchn, cpu);
 893         } else {
 894                 struct irq_info *info = info_for_irq(irq);
 895                 WARN_ON(info == NULL || info->type != IRQT_IPI);
 896         }
 897
 898  out:
 899         mutex_unlock(&irq_mapping_update_lock);
 900         return irq;
 901 }
 902
 903 int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
 904                                    unsigned int remote_port)
 905 {
 906         struct evtchn_bind_interdomain bind_interdomain;
 907         int err;
 908
 909         bind_interdomain.remote_dom  = remote_domain;
 910         bind_interdomain.remote_port = remote_port;
 911
 912         err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
 913                                           &bind_interdomain);
 914
 915         return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
 916 }
 917 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
 918
 919 static int find_virq(unsigned int virq, unsigned int cpu)
 920 {
 921         struct evtchn_status status;
 922         int port, rc = -ENOENT;
 923
 924         memset(&status, 0, sizeof(status));
 925         for (port = 0; port < xen_evtchn_max_channels(); port++) {
 926                 status.dom = DOMID_SELF;
 927                 status.port = port;
 928                 rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
 929                 if (rc < 0)
 930                         continue;
 931                 if (status.status != EVTCHNSTAT_virq)
 932                         continue;
 933                 if (status.u.virq == virq && status.vcpu == cpu) {
 934                         rc = port;
 935                         break;
 936                 }
 937         }
 938         return rc;
 939 }
 940
 941 /**
 942  * xen_evtchn_nr_channels - number of usable event channel ports
 943  *
 944  * This may be less than the maximum supported by the current
 945  * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
 946  * supported.
 947  */
 948 unsigned xen_evtchn_nr_channels(void)
 949 {
 950         return evtchn_ops->nr_channels();
 951 }
 952 EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
 953
 954 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
 955 {
 956         struct evtchn_bind_virq bind_virq;
 957         int evtchn, irq, ret;
 958
 959         mutex_lock(&irq_mapping_update_lock);
 960
 961         irq = per_cpu(virq_to_irq, cpu)[virq];
 962
 963         if (irq == -1) {
 964                 irq = xen_allocate_irq_dynamic();
 965                 if (irq < 0)
 966                         goto out;
 967
 968                 irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 969                                               handle_percpu_irq, "virq");
 970
 971                 bind_virq.virq = virq;
 972                 bind_virq.vcpu = cpu;
 973                 ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 974                                                 &bind_virq);
 975                 if (ret == 0)
 976                         evtchn = bind_virq.port;
 977                 else {
 978                         if (ret == -EEXIST)
 979                                 ret = find_virq(virq, cpu);
 980                         BUG_ON(ret < 0);
 981                         evtchn = ret;
 982                 }
 983
 984                 ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
 985                 if (ret < 0) {
 986                         __unbind_from_irq(irq);
 987                         irq = ret;
 988                         goto out;
 989                 }
 990
 991                 bind_evtchn_to_cpu(evtchn, cpu);
 992         } else {
 993                 struct irq_info *info = info_for_irq(irq);
 994                 WARN_ON(info == NULL || info->type != IRQT_VIRQ);
 995         }
 996
 997 out:
 998         mutex_unlock(&irq_mapping_update_lock);
 999
1000         return irq;
1001 }
1002
1003 static void unbind_from_irq(unsigned int irq)
1004 {
1005         mutex_lock(&irq_mapping_update_lock);
1006         __unbind_from_irq(irq);
1007         mutex_unlock(&irq_mapping_update_lock);
1008 }
1009
1010 int bind_evtchn_to_irqhandler(unsigned int evtchn,
1011                               irq_handler_t handler,
1012                               unsigned long irqflags,
1013                               const char *devname, void *dev_id)
1014 {
1015         int irq, retval;
1016
1017         irq = bind_evtchn_to_irq(evtchn);
1018         if (irq < 0)
1019                 return irq;
1020         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1021         if (retval != 0) {
1022                 unbind_from_irq(irq);
1023                 return retval;
1024         }
1025
1026         return irq;
1027 }
1028 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1029
1030 int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1031                                           unsigned int remote_port,
1032                                           irq_handler_t handler,
1033                                           unsigned long irqflags,
1034                                           const char *devname,
1035                                           void *dev_id)
1036 {
1037         int irq, retval;
1038
1039         irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1040         if (irq < 0)
1041                 return irq;
1042
1043         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1044         if (retval != 0) {
1045                 unbind_from_irq(irq);
1046                 return retval;
1047         }
1048
1049         return irq;
1050 }
1051 EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1052
1053 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1054                             irq_handler_t handler,
1055                             unsigned long irqflags, const char *devname, void *dev_id)
1056 {
1057         int irq, retval;
1058
1059         irq = bind_virq_to_irq(virq, cpu);
1060         if (irq < 0)
1061                 return irq;
1062         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1063         if (retval != 0) {
1064                 unbind_from_irq(irq);
1065                 return retval;
1066         }
1067
1068         return irq;
1069 }
1070 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1071
1072 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1073                            unsigned int cpu,
1074                            irq_handler_t handler,
1075                            unsigned long irqflags,
1076                            const char *devname,
1077                            void *dev_id)
1078 {
1079         int irq, retval;
1080
1081         irq = bind_ipi_to_irq(ipi, cpu);
1082         if (irq < 0)
1083                 return irq;
1084
1085         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1086         retval = request_irq(irq, handler, irqflags, devname, dev_id);
1087         if (retval != 0) {
1088                 unbind_from_irq(irq);
1089                 return retval;
1090         }
1091
1092         return irq;
1093 }
1094
1095 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1096 {
1097         struct irq_info *info = irq_get_handler_data(irq);
1098
1099         if (WARN_ON(!info))
1100                 return;
1101         free_irq(irq, dev_id);
1102         unbind_from_irq(irq);
1103 }
1104 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1105
1106 /**
1107  * xen_set_irq_priority() - set an event channel priority.
1108  * @irq:irq bound to an event channel.
1109  * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1110  */
1111 int xen_set_irq_priority(unsigned irq, unsigned priority)
1112 {
1113         struct evtchn_set_priority set_priority;
1114
1115         set_priority.port = evtchn_from_irq(irq);
1116         set_priority.priority = priority;
1117
1118         return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1119                                            &set_priority);
1120 }
1121 EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1122
1123 int evtchn_make_refcounted(unsigned int evtchn)
1124 {
1125         int irq = get_evtchn_to_irq(evtchn);
1126         struct irq_info *info;
1127
1128         if (irq == -1)
1129                 return -ENOENT;
1130
1131         info = irq_get_handler_data(irq);
1132
1133         if (!info)
1134                 return -ENOENT;
1135
1136         WARN_ON(info->refcnt != -1);
1137
1138         info->refcnt = 1;
1139
1140         return 0;
1141 }
1142 EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1143
1144 int evtchn_get(unsigned int evtchn)
1145 {
1146         int irq;
1147         struct irq_info *info;
1148         int err = -ENOENT;
1149
1150         if (evtchn >= xen_evtchn_max_channels())
1151                 return -EINVAL;
1152
1153         mutex_lock(&irq_mapping_update_lock);
1154
1155         irq = get_evtchn_to_irq(evtchn);
1156         if (irq == -1)
1157                 goto done;
1158
1159         info = irq_get_handler_data(irq);
1160
1161         if (!info)
1162                 goto done;
1163
1164         err = -EINVAL;
1165         if (info->refcnt <= 0)
1166                 goto done;
1167
1168         info->refcnt++;
1169         err = 0;
1170  done:
1171         mutex_unlock(&irq_mapping_update_lock);
1172
1173         return err;
1174 }
1175 EXPORT_SYMBOL_GPL(evtchn_get);
1176
1177 void evtchn_put(unsigned int evtchn)
1178 {
1179         int irq = get_evtchn_to_irq(evtchn);
1180         if (WARN_ON(irq == -1))
1181                 return;
1182         unbind_from_irq(irq);
1183 }
1184 EXPORT_SYMBOL_GPL(evtchn_put);
1185
1186 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1187 {
1188         int irq;
1189
1190 #ifdef CONFIG_X86
1191         if (unlikely(vector == XEN_NMI_VECTOR)) {
1192                 int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, cpu, NULL);
1193                 if (rc < 0)
1194                         printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1195                 return;
1196         }
1197 #endif
1198         irq = per_cpu(ipi_to_irq, cpu)[vector];
1199         BUG_ON(irq < 0);
1200         notify_remote_via_irq(irq);
1201 }
1202
1203 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1204
1205 static void __xen_evtchn_do_upcall(void)
1206 {
1207         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1208         int cpu = get_cpu();
1209         unsigned count;
1210
1211         do {
1212                 vcpu_info->evtchn_upcall_pending = 0;
1213
1214                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1215                         goto out;
1216
1217                 xen_evtchn_handle_events(cpu);
1218
1219                 BUG_ON(!irqs_disabled());
1220
1221                 count = __this_cpu_read(xed_nesting_count);
1222                 __this_cpu_write(xed_nesting_count, 0);
1223         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1224
1225 out:
1226
1227         put_cpu();
1228 }
1229
1230 void xen_evtchn_do_upcall(struct pt_regs *regs)
1231 {
1232         struct pt_regs *old_regs = set_irq_regs(regs);
1233
1234         irq_enter();
1235 #ifdef CONFIG_X86
1236         exit_idle();
1237         inc_irq_stat(irq_hv_callback_count);
1238 #endif
1239
1240         __xen_evtchn_do_upcall();
1241
1242         irq_exit();
1243         set_irq_regs(old_regs);
1244 }
1245
1246 void xen_hvm_evtchn_do_upcall(void)
1247 {
1248         __xen_evtchn_do_upcall();
1249 }
1250 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1251
1252 /* Rebind a new event channel to an existing irq. */
1253 void rebind_evtchn_irq(int evtchn, int irq)
1254 {
1255         struct irq_info *info = info_for_irq(irq);
1256
1257         if (WARN_ON(!info))
1258                 return;
1259
1260         /* Make sure the irq is masked, since the new event channel
1261            will also be masked. */
1262         disable_irq(irq);
1263
1264         mutex_lock(&irq_mapping_update_lock);
1265
1266         /* After resume the irq<->evtchn mappings are all cleared out */
1267         BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1268         /* Expect irq to have been bound before,
1269            so there should be a proper type */
1270         BUG_ON(info->type == IRQT_UNBOUND);
1271
1272         (void)xen_irq_info_evtchn_setup(irq, evtchn);
1273
1274         mutex_unlock(&irq_mapping_update_lock);
1275
1276         /* new event channels are always bound to cpu 0 */
1277         irq_set_affinity(irq, cpumask_of(0));
1278
1279         /* Unmask the event channel. */
1280         enable_irq(irq);
1281 }
1282
1283 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1284 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1285 {
1286         struct evtchn_bind_vcpu bind_vcpu;
1287         int evtchn = evtchn_from_irq(irq);
1288         int masked;
1289
1290         if (!VALID_EVTCHN(evtchn))
1291                 return -1;
1292
1293         /*
1294          * Events delivered via platform PCI interrupts are always
1295          * routed to vcpu 0 and hence cannot be rebound.
1296          */
1297         if (xen_hvm_domain() && !xen_have_vector_callback)
1298                 return -1;
1299
1300         /* Send future instances of this interrupt to other vcpu. */
1301         bind_vcpu.port = evtchn;
1302         bind_vcpu.vcpu = tcpu;
1303
1304         /*
1305          * Mask the event while changing the VCPU binding to prevent
1306          * it being delivered on an unexpected VCPU.
1307          */
1308         masked = test_and_set_mask(evtchn);
1309
1310         /*
1311          * If this fails, it usually just indicates that we're dealing with a
1312          * virq or IPI channel, which don't actually need to be rebound. Ignore
1313          * it, but don't do the xenlinux-level rebind in that case.
1314          */
1315         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1316                 bind_evtchn_to_cpu(evtchn, tcpu);
1317
1318         if (!masked)
1319                 unmask_evtchn(evtchn);
1320
1321         return 0;
1322 }
1323
1324 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1325                             bool force)
1326 {
1327         unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1328
1329         return rebind_irq_to_cpu(data->irq, tcpu);
1330 }
1331
1332 static void enable_dynirq(struct irq_data *data)
1333 {
1334         int evtchn = evtchn_from_irq(data->irq);
1335
1336         if (VALID_EVTCHN(evtchn))
1337                 unmask_evtchn(evtchn);
1338 }
1339
1340 static void disable_dynirq(struct irq_data *data)
1341 {
1342         int evtchn = evtchn_from_irq(data->irq);
1343
1344         if (VALID_EVTCHN(evtchn))
1345                 mask_evtchn(evtchn);
1346 }
1347
1348 static void ack_dynirq(struct irq_data *data)
1349 {
1350         int evtchn = evtchn_from_irq(data->irq);
1351
1352         irq_move_irq(data);
1353
1354         if (VALID_EVTCHN(evtchn))
1355                 clear_evtchn(evtchn);
1356 }
1357
1358 static void mask_ack_dynirq(struct irq_data *data)
1359 {
1360         disable_dynirq(data);
1361         ack_dynirq(data);
1362 }
1363
1364 static int retrigger_dynirq(struct irq_data *data)
1365 {
1366         unsigned int evtchn = evtchn_from_irq(data->irq);
1367         int masked;
1368
1369         if (!VALID_EVTCHN(evtchn))
1370                 return 0;
1371
1372         masked = test_and_set_mask(evtchn);
1373         set_evtchn(evtchn);
1374         if (!masked)
1375                 unmask_evtchn(evtchn);
1376
1377         return 1;
1378 }
1379
1380 static void restore_pirqs(void)
1381 {
1382         int pirq, rc, irq, gsi;
1383         struct physdev_map_pirq map_irq;
1384         struct irq_info *info;
1385
1386         list_for_each_entry(info, &xen_irq_list_head, list) {
1387                 if (info->type != IRQT_PIRQ)
1388                         continue;
1389
1390                 pirq = info->u.pirq.pirq;
1391                 gsi = info->u.pirq.gsi;
1392                 irq = info->irq;
1393
1394                 /* save/restore of PT devices doesn't work, so at this point the
1395                  * only devices present are GSI based emulated devices */
1396                 if (!gsi)
1397                         continue;
1398
1399                 map_irq.domid = DOMID_SELF;
1400                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1401                 map_irq.index = gsi;
1402                 map_irq.pirq = pirq;
1403
1404                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1405                 if (rc) {
1406                         pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1407                                 gsi, irq, pirq, rc);
1408                         xen_free_irq(irq);
1409                         continue;
1410                 }
1411
1412                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1413
1414                 __startup_pirq(irq);
1415         }
1416 }
1417
1418 static void restore_cpu_virqs(unsigned int cpu)
1419 {
1420         struct evtchn_bind_virq bind_virq;
1421         int virq, irq, evtchn;
1422
1423         for (virq = 0; virq < NR_VIRQS; virq++) {
1424                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1425                         continue;
1426
1427                 BUG_ON(virq_from_irq(irq) != virq);
1428
1429                 /* Get a new binding from Xen. */
1430                 bind_virq.virq = virq;
1431                 bind_virq.vcpu = cpu;
1432                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1433                                                 &bind_virq) != 0)
1434                         BUG();
1435                 evtchn = bind_virq.port;
1436
1437                 /* Record the new mapping. */
1438                 (void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1439                 bind_evtchn_to_cpu(evtchn, cpu);
1440         }
1441 }
1442
1443 static void restore_cpu_ipis(unsigned int cpu)
1444 {
1445         struct evtchn_bind_ipi bind_ipi;
1446         int ipi, irq, evtchn;
1447
1448         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1449                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1450                         continue;
1451
1452                 BUG_ON(ipi_from_irq(irq) != ipi);
1453
1454                 /* Get a new binding from Xen. */
1455                 bind_ipi.vcpu = cpu;
1456                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1457                                                 &bind_ipi) != 0)
1458                         BUG();
1459                 evtchn = bind_ipi.port;
1460
1461                 /* Record the new mapping. */
1462                 (void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1463                 bind_evtchn_to_cpu(evtchn, cpu);
1464         }
1465 }
1466
1467 /* Clear an irq's pending state, in preparation for polling on it */
1468 void xen_clear_irq_pending(int irq)
1469 {
1470         int evtchn = evtchn_from_irq(irq);
1471
1472         if (VALID_EVTCHN(evtchn))
1473                 clear_evtchn(evtchn);
1474 }
1475 EXPORT_SYMBOL(xen_clear_irq_pending);
1476 void xen_set_irq_pending(int irq)
1477 {
1478         int evtchn = evtchn_from_irq(irq);
1479
1480         if (VALID_EVTCHN(evtchn))
1481                 set_evtchn(evtchn);
1482 }
1483
1484 bool xen_test_irq_pending(int irq)
1485 {
1486         int evtchn = evtchn_from_irq(irq);
1487         bool ret = false;
1488
1489         if (VALID_EVTCHN(evtchn))
1490                 ret = test_evtchn(evtchn);
1491
1492         return ret;
1493 }
1494
1495 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1496  * the irq will be disabled so it won't deliver an interrupt. */
1497 void xen_poll_irq_timeout(int irq, u64 timeout)
1498 {
1499         evtchn_port_t evtchn = evtchn_from_irq(irq);
1500
1501         if (VALID_EVTCHN(evtchn)) {
1502                 struct sched_poll poll;
1503
1504                 poll.nr_ports = 1;
1505                 poll.timeout = timeout;
1506                 set_xen_guest_handle(poll.ports, &evtchn);
1507
1508                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1509                         BUG();
1510         }
1511 }
1512 EXPORT_SYMBOL(xen_poll_irq_timeout);
1513 /* Poll waiting for an irq to become pending.  In the usual case, the
1514  * irq will be disabled so it won't deliver an interrupt. */
1515 void xen_poll_irq(int irq)
1516 {
1517         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1518 }
1519
1520 /* Check whether the IRQ line is shared with other guests. */
1521 int xen_test_irq_shared(int irq)
1522 {
1523         struct irq_info *info = info_for_irq(irq);
1524         struct physdev_irq_status_query irq_status;
1525
1526         if (WARN_ON(!info))
1527                 return -ENOENT;
1528
1529         irq_status.irq = info->u.pirq.pirq;
1530
1531         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1532                 return 0;
1533         return !(irq_status.flags & XENIRQSTAT_shared);
1534 }
1535 EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1536
1537 void xen_irq_resume(void)
1538 {
1539         unsigned int cpu;
1540         struct irq_info *info;
1541
1542         /* New event-channel space is not 'live' yet. */
1543         xen_evtchn_mask_all();
1544         xen_evtchn_resume();
1545
1546         /* No IRQ <-> event-channel mappings. */
1547         list_for_each_entry(info, &xen_irq_list_head, list)
1548                 info->evtchn = 0; /* zap event-channel binding */
1549
1550         clear_evtchn_to_irq_all();
1551
1552         for_each_possible_cpu(cpu) {
1553                 restore_cpu_virqs(cpu);
1554                 restore_cpu_ipis(cpu);
1555         }
1556
1557         restore_pirqs();
1558 }
1559
1560 static struct irq_chip xen_dynamic_chip __read_mostly = {
1561         .name                   = "xen-dyn",
1562
1563         .irq_disable            = disable_dynirq,
1564         .irq_mask               = disable_dynirq,
1565         .irq_unmask             = enable_dynirq,
1566
1567         .irq_ack                = ack_dynirq,
1568         .irq_mask_ack           = mask_ack_dynirq,
1569
1570         .irq_set_affinity       = set_affinity_irq,
1571         .irq_retrigger          = retrigger_dynirq,
1572 };
1573
1574 static struct irq_chip xen_pirq_chip __read_mostly = {
1575         .name                   = "xen-pirq",
1576
1577         .irq_startup            = startup_pirq,
1578         .irq_shutdown           = shutdown_pirq,
1579         .irq_enable             = enable_pirq,
1580         .irq_disable            = disable_pirq,
1581
1582         .irq_mask               = disable_dynirq,
1583         .irq_unmask             = enable_dynirq,
1584
1585         .irq_ack                = eoi_pirq,
1586         .irq_eoi                = eoi_pirq,
1587         .irq_mask_ack           = mask_ack_pirq,
1588
1589         .irq_set_affinity       = set_affinity_irq,
1590
1591         .irq_retrigger          = retrigger_dynirq,
1592 };
1593
1594 static struct irq_chip xen_percpu_chip __read_mostly = {
1595         .name                   = "xen-percpu",
1596
1597         .irq_disable            = disable_dynirq,
1598         .irq_mask               = disable_dynirq,
1599         .irq_unmask             = enable_dynirq,
1600
1601         .irq_ack                = ack_dynirq,
1602 };
1603
1604 int xen_set_callback_via(uint64_t via)
1605 {
1606         struct xen_hvm_param a;
1607         a.domid = DOMID_SELF;
1608         a.index = HVM_PARAM_CALLBACK_IRQ;
1609         a.value = via;
1610         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1611 }
1612 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1613
1614 #ifdef CONFIG_XEN_PVHVM
1615 /* Vector callbacks are better than PCI interrupts to receive event
1616  * channel notifications because we can receive vector callbacks on any
1617  * vcpu and we don't need PCI support or APIC interactions. */
1618 void xen_callback_vector(void)
1619 {
1620         int rc;
1621         uint64_t callback_via;
1622         if (xen_have_vector_callback) {
1623                 callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1624                 rc = xen_set_callback_via(callback_via);
1625                 if (rc) {
1626                         pr_err("Request for Xen HVM callback vector failed\n");
1627                         xen_have_vector_callback = 0;
1628                         return;
1629                 }
1630                 pr_info("Xen HVM callback vector for event delivery is enabled\n");
1631                 /* in the restore case the vector has already been allocated */
1632                 if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1633                         alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1634                                         xen_hvm_callback_vector);
1635         }
1636 }
1637 #else
1638 void xen_callback_vector(void) {}
1639 #endif
1640
1641 #undef MODULE_PARAM_PREFIX
1642 #define MODULE_PARAM_PREFIX "xen."
1643
1644 static bool fifo_events = true;
1645 module_param(fifo_events, bool, 0);
1646
1647 void __init xen_init_IRQ(void)
1648 {
1649         int ret = -EINVAL;
1650
1651         if (fifo_events)
1652                 ret = xen_evtchn_fifo_init();
1653         if (ret < 0)
1654                 xen_evtchn_2l_init();
1655
1656         evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1657                                 sizeof(*evtchn_to_irq), GFP_KERNEL);
1658         BUG_ON(!evtchn_to_irq);
1659
1660         /* No event channels are 'live' right now. */
1661         xen_evtchn_mask_all();
1662
1663         pirq_needs_eoi = pirq_needs_eoi_flag;
1664
1665 #ifdef CONFIG_X86
1666         if (xen_pv_domain()) {
1667                 irq_ctx_init(smp_processor_id());
1668                 if (xen_initial_domain())
1669                         pci_xen_initial_domain();
1670         }
1671         if (xen_feature(XENFEAT_hvm_callback_vector))
1672                 xen_callback_vector();
1673
1674         if (xen_hvm_domain()) {
1675                 native_init_IRQ();
1676                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1677                  * __acpi_register_gsi can point at the right function */
1678                 pci_xen_hvm_init();
1679         } else {
1680                 int rc;
1681                 struct physdev_pirq_eoi_gmfn eoi_gmfn;
1682
1683                 pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1684                 eoi_gmfn.gmfn = virt_to_mfn(pirq_eoi_map);
1685                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1686                 /* TODO: No PVH support for PIRQ EOI */
1687                 if (rc != 0) {
1688                         free_page((unsigned long) pirq_eoi_map);
1689                         pirq_eoi_map = NULL;
1690                 } else
1691                         pirq_needs_eoi = pirq_check_eoi_map;
1692         }
1693 #endif
1694 }